Morbidity and Mortality 2014-2019 Analysis Report (Prior to Covid-19)
Data Analysis in R Project (Academic)
by Patrick Cook
5/27/2021
Introduction
This report explores the Center for Disease Control’s (CDC) Weekly Morbidity and Mortality data from 2014 to 2019. The data includes weekly death counts for specific causes in the United States. The data also includes weekly death counts for individual states and specific areas such as New York City, Puerto Rico and Washington, D.C. The data is voluntary and not guaranteed to be reported on a regular basis. Therefore, inference and causality statements are based on this data set and may not apply to the country or state as a whole. Information on the data set can be found in the data description file included in the data folder.
Data Structure Exploration
The data set contains 16,902 observations and 30 variables with the names shown below.
## [1] "Jurisdiction.of.Occurrence"
## [2] "MMWR.Year"
## [3] "MMWR.Week"
## [4] "Week.Ending.Date"
## [5] "All..Cause"
## [6] "Natural.Cause"
## [7] "Septicemia..A40.A41."
## [8] "Malignant.neoplasms..C00.C97."
## [9] "Diabetes.mellitus..E10.E14."
## [10] "Alzheimer.disease..G30."
## [11] "Influenza.and.pneumonia..J10.J18."
## [12] "Chronic.lower.respiratory.diseases..J40.J47."
## [13] "Other.diseases.of.respiratory.system..J00.J06.J30.J39.J67.J70.J98."
## [14] "Nephritis..nephrotic.syndrome.and.nephrosis..N00.N07.N17.N19.N25.N27."
## [15] "Symptoms..signs.and.abnormal.clinical.and.laboratory.findings..not.elsewhere.classified..R00.R99."
## [16] "Diseases.of.heart..I00.I09.I11.I13.I20.I51."
## [17] "Cerebrovascular.diseases..I60.I69."
## [18] "flag_allcause"
## [19] "flag_natcause"
## [20] "flag_sept"
## [21] "flag_neopl"
## [22] "flag_diab"
## [23] "flag_alz"
## [24] "flag_inflpn"
## [25] "flag_clrd"
## [26] "flag_otherresp"
## [27] "flag_nephr"
## [28] "flag_otherunk"
## [29] "flag_hd"
## [30] "flag_stroke"The names given for the causes of death include clinical information and codes. These names will be shortened to the primary organ involved or a two word descriptor.
Of the 30 columns, 13 are used as flags indicating suppressed death data for weekly counts between the values of 1 and 9. These values are suppressed in the original data to protect privacy. The flag columns will be dropped and the corresponding missing values for the causes of death will not be removed or replaced.
The reasoning for not dropping these rows is that the rows contain other useful information on other individual death causes. In addition, the missing values do give information such as the value was not zero but a value between 1 and 9. The missing values are not replaced due to the affect these artificial values would have on the data leading to false assumptions and/or conclusions.
A row named “week_sum” was added, giving the sum of all weekly causes of death except “all_causes” and “natural”. These variables include counts given in other columns and therefore are excluded from the added sum column to prevent double counting.
After the changes, the new column names are:
## [1] "d_location" "d_year" "d_week" "ending_week"
## [5] "all_causes" "natural" "sepsis" "malignant"
## [9] "diabetes" "alzheimer" "flu_pneum" "lower_resp"
## [13] "other_resp" "kidney" "abnormal_finding" "heart"
## [17] "brain" "week_sum"The d_location, d_year and d_week columns hold categorical information and will be changed to factors. Ending_week gives the date of the last day for the week counted and will be changed into a date. There is a 53rd week for the 2014 year. All weeks will be shifted back 1 so that week 53 in 2014 becomes 52, 52 becomes 51, etc… The new week 0 will be dropped. The new structure of the data is shown below.
The data set now contains 18 columns and 16,848 observations.
## 'data.frame': 16848 obs. of 18 variables:
## $ d_location : Factor w/ 54 levels "Alabama","Alaska",..: 47 47 47 47 47 47 47 47 47 47 ...
## $ d_year : Ord.factor w/ 6 levels "2014"<"2015"<..: 1 1 1 1 1 1 1 1 1 1 ...
## $ d_week : Ord.factor w/ 52 levels "1"<"2"<"3"<"4"<..: 1 2 3 4 5 6 7 8 9 10 ...
## $ ending_week : Date, format: "2014-01-11" "2014-01-18" ...
## $ all_causes : int 56353 54769 54223 54074 53484 53910 52752 51768 52163 51912 ...
## $ natural : int 52450 51043 50560 50402 49790 50175 49010 47907 48353 48111 ...
## $ sepsis : int 905 919 845 890 849 851 708 779 777 819 ...
## $ malignant : int 11504 11496 11629 11584 11355 11477 11478 11251 11535 11340 ...
## $ diabetes : int 1735 1660 1602 1586 1643 1642 1564 1588 1536 1584 ...
## $ alzheimer : int 1917 1914 1862 1867 1873 1843 1814 1776 1830 1775 ...
## $ flu_pneum : int 1910 1920 1765 1642 1528 1472 1269 1228 1215 1153 ...
## $ lower_resp : int 3444 3333 3467 3283 3351 3303 3047 3008 3043 3112 ...
## $ other_resp : int 845 812 753 720 728 739 731 687 760 730 ...
## $ kidney : int 1098 1056 1029 998 1038 1021 973 1018 1040 987 ...
## $ abnormal_finding: int 665 598 622 664 641 638 643 595 642 611 ...
## $ heart : int 13663 12928 12813 12896 12681 12984 12577 12248 12318 12375 ...
## $ brain : int 2738 2714 2720 2699 2684 2669 2799 2630 2529 2555 ...
## $ week_sum : int 40424 39350 39107 38829 38371 38639 37603 36808 37225 37041 ...The structure of the new data frame after the changes is shown above.
Analysis
Distribution of US Data
Next, the data is split into a data frame for the United States data and another containing the state and territory data. The US subset contains 312 observations and the State subset contains 16,536 observations. Next, we will look at summary statistics and missing data in both data subsets.
## d_location d_year d_week ending_week all_causes natural
## United States:312 2014:52 1 : 6 Min. :2014-01-11 Min. :46751 Min. :42681
## 2015:52 2 : 6 1st Qu.:2015-07-09 1st Qu.:50453 1st Qu.:45849
## 2016:52 3 : 6 Median :2017-01-03 Median :52223 Median :47731
## 2017:52 4 : 6 Mean :2017-01-03 Mean :52994 Mean :48526
## 2018:52 5 : 6 3rd Qu.:2018-07-01 3rd Qu.:55467 3rd Qu.:50897
## 2019:52 6 : 6 Max. :2019-12-28 Max. :67501 Max. :62573
## (Other):276
## sepsis malignant diabetes alzheimer flu_pneum lower_resp
## Min. : 629.0 Min. :10877 Min. :1285 Min. :1498 Min. : 625.0 Min. :2310
## 1st Qu.: 712.8 1st Qu.:11302 1st Qu.:1465 1st Qu.:2031 1st Qu.: 781.0 1st Qu.:2654
## Median : 747.5 Median :11430 Median :1543 Median :2178 Median : 882.5 Median :2889
## Mean : 767.8 Mean :11450 Mean :1572 Mean :2189 Mean :1047.6 Mean :2981
## 3rd Qu.: 817.2 3rd Qu.:11593 3rd Qu.:1664 3rd Qu.:2381 3rd Qu.:1217.0 3rd Qu.:3304
## Max. :1066.0 Max. :12408 Max. :2110 Max. :3212 Max. :2930.0 Max. :4373
##
## other_resp kidney abnormal_finding heart brain week_sum
## Min. : 589.0 Min. : 796.0 Min. :494.0 Min. :10497 Min. :2216 Min. :32493
## 1st Qu.: 712.0 1st Qu.: 899.0 1st Qu.:587.0 1st Qu.:11527 1st Qu.:2591 1st Qu.:35296
## Median : 768.0 Median : 954.0 Median :615.5 Median :12096 Median :2718 Median :36771
## Mean : 777.1 Mean : 963.9 Mean :623.9 Mean :12281 Mean :2747 Mean :37399
## 3rd Qu.: 837.0 3rd Qu.:1029.0 3rd Qu.:652.0 3rd Qu.:12979 3rd Qu.:2899 3rd Qu.:39161
## Max. :1086.0 Max. :1237.0 Max. :827.0 Max. :16020 Max. :3364 Max. :48294
## 
Most weekly death categories look normally distributed with varying levels of right skew. The US data has no missing values. Next, is an examination of state distributions and missing data.
State Data Distribution
## d_location d_year d_week ending_week all_causes natural
## Alabama : 312 2014:2756 1 : 318 Min. :2014-01-11 Min. : 56 Min. : 45.0
## Alaska : 312 2015:2756 2 : 318 1st Qu.:2015-07-09 1st Qu.: 304 1st Qu.: 273.0
## Arizona : 312 2016:2756 3 : 318 Median :2017-01-03 Median : 766 Median : 693.0
## Arkansas : 312 2017:2756 4 : 318 Mean :2017-01-03 Mean :1011 Mean : 925.5
## California: 312 2018:2756 5 : 318 3rd Qu.:2018-07-01 3rd Qu.:1256 3rd Qu.:1154.0
## Colorado : 312 2019:2756 6 : 318 Max. :2019-12-28 Max. :7009 Max. :6510.0
## (Other) :14664 (Other):14628
## sepsis malignant diabetes alzheimer flu_pneum lower_resp
## Min. : 0.00 Min. : 10 Min. : 0.00 Min. : 0.00 Min. : 0.00 Min. : 0.00
## 1st Qu.: 13.00 1st Qu.: 62 1st Qu.: 20.00 1st Qu.: 21.00 1st Qu.: 14.00 1st Qu.: 28.00
## Median : 20.00 Median : 162 Median : 30.00 Median : 38.00 Median : 21.00 Median : 49.00
## Mean : 23.23 Mean : 218 Mean : 39.55 Mean : 51.25 Mean : 28.84 Mean : 64.68
## 3rd Qu.: 31.00 3rd Qu.: 270 3rd Qu.: 46.00 3rd Qu.: 61.00 3rd Qu.: 36.00 3rd Qu.: 82.00
## Max. :133.00 Max. :1298 Max. :285.00 Max. :447.00 Max. :457.00 Max. :434.00
## NA's :7331 NA's :11 NA's :4171 NA's :3307 NA's :6028 NA's :2250
## other_resp kidney abnormal_finding heart brain week_sum
## Min. : 0.00 Min. : 0.0 Min. : 0.0 Min. : 10.0 Min. : 0.00 Min. : 10.0
## 1st Qu.:13.00 1st Qu.: 13.0 1st Qu.:11.0 1st Qu.: 63.0 1st Qu.: 24.00 1st Qu.: 176.0
## Median :19.00 Median : 22.0 Median :16.0 Median : 167.0 Median : 44.00 Median : 510.5
## Mean :22.93 Mean : 27.5 Mean :19.3 Mean : 233.9 Mean : 60.79 Mean : 700.1
## 3rd Qu.:30.00 3rd Qu.: 37.0 3rd Qu.:24.0 3rd Qu.: 298.0 3rd Qu.: 71.00 3rd Qu.: 882.0
## Max. :97.00 Max. :120.0 Max. :93.0 Max. :1672.0 Max. :423.00 Max. :5214.0
## NA's :7256 NA's :6494 NA's :8074 NA's :20 NA's :2557The state data has missing values for 11 of the variables. These are due to the privacy flags and are known to be some value between 1 and 9 deaths. The abnormal category followed by sepsis and other respiratory have the highest missing values as shown in the following figure.
The median and mean are similar for most causes of death for the US data set except for flu_pneum. State data has more significant differences in the median and mean value. The mean values are higher than the median showing right skewness. This will be confirmed during the exploration of the state data. It is also noted that for all the disease causes of deaths, only heart disease and malignant have non-zero minimums. This along with the NA information indicates that these two causes are responsible for at least 1 death in every state and territory each week.
The distributions above show large right skew for all graphs. There are gaps in the histograms from 0 to 10 for sepsis, other respiratory, kidney and abnormal findings. These are due to the privacy excluded data. Several graphs show multi-modal areas to the right. This may be due to differing populations of states.
All death categories except all_causes, natural and week_sum have missing values. Those categories being general categories are more likely to contain death counts higher than 9 per week. Abnormal findings being the highest shows that there are few (1 to 9) abnormal causes of death per week. It is interesting that heart and malignant causes have weeks where there are 1 to 9 deaths. These are the highest causes of death in the United States. Those are likely from smaller population states or territories. The next figure shows the states ordered by missing values.
The general pattern shows that states and regions with smaller populations have more causes of weekly deaths of 1 to 9. Whereas, more populous states have fewer death weekly counts of 1 to 9. Interestingly, Ohio only has one instance of a weekly death count being 1 to 9 deaths. When California, with the largest population, has 2 weeks with a cause of death being 1 to 9.
The next figure shows the US all_causes and week_sum categories chronologically.
US Chronoligical Data
The scatter plots show sinusoidal or seasonal pattern with peak deaths occurring in Winter, the beginning of the year. The number of deaths have a positive upward trend that is noticeable when looking at the minimums for the year.
The next figure examines what other diseases have a seasonality trend and which causes of weekly deaths have an upward year-over-year trend.
Based on the US data, all categories for causes of death are increasing except flu_pnuem and sepsis. It is interesting that flu and pneumonia deaths were decreasing prior to Covid-19. Notable increases are Alzheimer, brain disease and diabetes deaths.
Also notable, even though flu and pneumonia deaths are decreasing and lower respiratory shows only a small upward trend, other respiratory conditions seem to be increasing significantly. The severity of the 2018 (2018-2019) flu season seems to effect all categories of death except abnormal findings. The 2015 (2015-2016) flu season is also seen but to a lesser extent. It’s possible these non-flu weekly death increases were related to the additional stress the flu has on the immune system and body.
The next plot organizes the causes of death on the same scale to significance of each cause of death and long term trends in growth rates.
US Cause Categories Growth Trends
The above graphic shows that heart disease deaths are increasing the most for the 2014 through 2019 time period followed by Alzheimer’s, brain disease and diabetes. The above figure highlights the significance of the 2018-19 flu season showing peaks in the heart disease, malignant, lower respiratory and Alzheimer weekly death categories.
The next figure is smoothed to compare the cause of death categories to each other to highlight seasonal trends. Not the vertical axis is in a log base 10 scale.
Heart disease and cancer deaths account for many more deaths in the US than the other causes in the data. Heart disease and cancer are around 4 times more than the next leading cause of deaths, lower respiratory. Though, weekly deaths in the brain disease category seem to be on pace to overtake lower respiratory. Alzheimer and diabetes are following a similar long-term growth trend.
The seasonal variability is made clearer after using smoothing and a log scale. There are even small peaks in the malignant data, which does not seem like it would have a seasonal relationship. I wonder how strong the correlation is between the causes. This could be due to seasonal reporting swings, lowered immune systems in the winter months, susceptibility to virus in the winter months due to enclosed spaces, or some other seasonal environmental cause. The seasonality of theses diseases is a topic for further study.
The next figure examines the correlation of the causes of death for the US data.
Correlation Between Causes
The above figure shows possible relationships between each cause of death category based on the chronological trends. The strongest correlations occur in the heart disease column and the weakest correlations are in the abnormal findings column. The strongest relationship is between heart disease and lower respiratory. This could be due to lower respiratory infections causing lower oxygen levels in the blood causing an already weakened heart, having to work harder. It is noted that there are other cause of death pairs that have correlation coefficients above 0.90, all relating to the heart category, with the exception of the brain and Alzheimer pair. These are only possible correlations that require a scientific study to prove correlation and possibly identify a causal relationship.
The distributions are normal with most having a right skewness as noted previously in the summary statistics data. Cancer (malignant) and other respiratory disease shows little to no skewness based on the histograms shown above.
The next figure shows year-over-year changes in each category to identify long-term trends and which years had the most significant increases.
Year Over Year Statistical Trends
The above shows the differences in the descriptive statistics by cause year over year. An increase is seen for all causes, natural causes, week sum, heart disease, cerebral (brain) diseases, diabetes and other respiratory diseases for all years. The Alzheimer category increased from 2014 to 2017 and has flattened out since 2017. The remaining diseases have flattened out or are declining for the last two years (2018-2019).
Each category of cause of death shows outliers for 2018 with flu and pneumonia reaching more than 300% of the median. Most causes in 2018 show mean and median divergence with the mean being greater than the median, confirming the right skewness seen in the distribution.
Knowing the Covid-19 virus was spreading in the world the last quarter of 2019, next we will compare the the causes of death look for the last quarters of each year.
End of Year (4th Quarter) Statistical Trends
Viewing only the 4th quarter data, shows all categories except cancer (malignant), flu and pneumonia, and sepsis have an upward trend from 2018 to 2019. The categories of all causes, natural, heart disease, cerebral (brain) disease, Alzheimer, diabetes, kidney disease and other respiratory show clear upward trends since 2015 for the 4th quarter data.
This increase could be due to the baby boomer population bubble aging, increased stress and hypertension leading to lowered immune system response, or some other factor. It also could be a common trend that has occurred in years prior to 2014. A broader data set including more years would be needed to examine this upward trend further.
The next table shows the most statistically significant growths (one-sided) between the mean of 2019 quarters and the mean of all previous years’ quarters (2014-2018).
US Most Significant Changes by Causes and Quarters
The most significant increases in causes of death categories for the entire year of 2019 are:
- diabetes in the 2nd quarter,
- all causes, diabetes, heart diseases and natural causes for the 3rd quarter,
- other respiratory for the 4th quarter
The next figure shows data graphically. The green dashed lines represent the 1st and 2nd standard deviations for the 2014-2018 data to demonstrate the significance of the 2019 growth.
The above figure compares the summary statistics of years 2014-2018 to 2019 for the entire year (weeks 1 through 52) including the 68% (1 sd) and 95% (2 sd) significance levels.
Other respiratory and diabetes categories of deaths show the largest increases in means when comparing the entire year data. Cerebral (brain) has the 2nd largest increase. All three of these categories have some values in the 4th quartile above 2 standard deviations. Only flu and pneumonia, and sepsis categories have decreases in means.
The next set of figures compares the 1st quarter data of 2019 to the 1st quarter data of all other years (2014-2018).
The 1st quarter data above shows smaller weekly death counts and narrower ranges due to a smaller time frame (quarters) of the data being used.
Diabetes has the greatest increase in significance when compared to the combined 1st quarter data of all previous years (2014-2018). Sepsis has the biggest decrease in significance. Cerebral (brain) and other respiratory means are in the 4th quartile (top 25%) of the previous year’s weekly death counts.
Comparing the 2nd quarter 2019 data to the 2nd quarter data of all other years shows increased means in all categories except flu and pneumonia, sepsis and abnormal findings.
Comparing the 2nd quarter 2019 data to the previous figure’s 1st quarter (weeks 1 to 13) data of 2019 shows increases in significance for all categories except cancer (malignant) and lower respiratory causes. The all causes, natural, week sum, heart cerebral (brain), diabetes and other respiratory are above one standard deviation (68%) of the previous years mean.
Diabetes is near 2 standard deviations of the previous years mean for the 2nd quarter data indicating a significant growth in U.S. deaths related to diabetes.
Comparing the means of the 3rd quarter data for 2019 to the means of the previous years’ 3rd quarter data shows increases in the means for all categories except the flu and pneumonia, and sepsis categories. The all causes, natural, heart, cerebral (brain), diabetes and other respiratory categories are still greater than 1 standard deviation with diabetes between 1 and 2 standard deviations.
Comparing the 3rd quarter 2019 data to the previous figure’s 2019 2nd quarter data shows decreases in significance levels for the categories: cancer, diabetes, flu and pneumonia, and sepsis. The 3rd quarter 2019 categories all causes, natural, week sum, heart, lower respiratory, cerebral (brain), Alzheimer, other respiratory and abnormal findings had increases when compared to the 2nd quarter of 2019.
Comparing the means of the 4th quarter data for 2019 to the means of the previous years’ 3rd quarter data shows decreases in means for the categories flu and pneumonia, sepsis and abnormal findings. Only the cerebral (brain), diabetes and other respiratory are still above 1 standard deviation of the previous years mean. Decreases in significance levels are seen for all categories except flu and pneumonia, and sepsis. Though, flu and pneumonia, and sepsis significance level did increase from the previous quarter, they are both still below previous years medians and means.
Summary of weekly death significance levels
Based on the previous 4 figures, there seems to be significant increases in deaths during the 2nd and 3rd quarters of 2019 when compared to previous years data indicating more deaths occurred in the spring and summer months than in previous years. Diabetes, cerebral (brain) and other respiratory categories show the greatest significant growth in deaths for these quarters. It is also noted that the flu and pneumonia deaths remained below the mean of previous years for all four quarters of 2019, possibly indicating a milder flu season for 2019.
US Cause of Death Categories Year over Year Comparisons
The areas between each years’ curves in the figures above shows the increases and decreases in weekly deaths. If the year is above the previous year, there is a increase in weekly deaths. If the year is below the previous year, there is a decrease in weekly deaths. The dashed lines separate the 2nd and 3rd quarters (spring and summer) from the 1st and 4th quarters (winter and fall).
For diabetes, the the 2019 curve (yellow) starts below the 2018 curve (blue) indicating less weekly diabetes deaths. Around week 6 the 2019 curve goes above the 2018 curve indicating more diabetes weekly deaths. The difference increases after week six until around week 32 where the deaths remain above the 2018 weekly deaths but are not as significant of a growth.
Cerebral (brain) and other respiratory deaths for 2019 show less significant growth than diabetes. For diabetes, cerebral (brain) and other respiratory, year 2014 is much lower than the other years and will result in lowering the aggregate means of the 2014-2018 data. This is especially true for the cerebral (brain) category. The minimums occur around week 30 and are between weeks 25 and 35 (Summer) for the top three causes. The maximums occur in the tail ends of the curves (stat of winter) for all categories. This seasonality seems reasonable for viral and bacterial diseases. For diabetes and cerebral deaths, it indicates a possible viral or bacterial connection or, more likely, a hidden relationship with other variables not present in the data set.
The curves for flu and pneumonia have been fairly consistent during the 2nd and 3rd quarters with larger differences observed at the tails during the 1st and 4th quarters (colder months). Both flu and pneumonia, and sepsis in 2019 fall within the range of previous years. These diseases do have lower weekly deaths than all previous years during the 3rd quarter of 2019.
The following figures examine the influence 2014 data has on lowering the mean for 2014-2018 aggregated data during the 2nd and 3rd quarters.
The figures above compares the 2014-2018, 2015-2018 and 2019 summary data for each cause of death. Removing the year 2014 from the summary data did raise the means for each cause of death category except flu and pneumonia. Though, the increases in the 1st and 2nd standard deviation targets were small. Cerebral (brain) and Alzheimer’s 2nd standard deviation even decreased slightly.
The U.S. data shows significant increases in weekly deaths for natural causes, diabetes, heart disease, cerebral (brain), natural causes and other respiratory. State and special regions will be examined next to identify if these increases are a general national trend or isolated to specific states or special regions.
State Data Analysis
Missing Data Analysis
The next figure examines the state and special region missing data.
The above figure shows the missing counts for each state including percent of missing data by cause of death category. The states are ordered from greatest population (California) to the smallest population (Wyoming). These missing values represent weekly death counts that are from 1 to 9. Missing states such as Texas indicate that the state had no weekly death counts from 1 to 9. Missing causes such as all causes and natural causes indicate there were no weeks with death counts between 1 and 9 for these categories.
As expected, states with smaller populations have more missing weekly death counts indicating they had more weekly death counts in the 1 to 9 range compared to larger states. This indicates that the more populous states have a greater influence on the state-level medians and means. The figure above shows interesting anomalies occurring with the states of Alabama, Tennessee, Massachusetts and New Jersey. These four states have much fewer weekly death cases from 1 to 9 compared to states with similar populations. This could be due to zero deaths occurring more often or more than 9 deaths occurring more often.
The most common cause of 1 to 9 weekly deaths is abnormal findings followed by sepsis, other respiratory and kidney. This pattern follows the magnitudes of weekly death categories seen in the US data when ordered from least to greatest. It is interesting that 6 of the 10 most populous states (California to Michigan) had weeks with weekly death counts from 1 to 9. Abnormal findings is the greatest contributor of missing weekly death counts for 6 of the 10 largest states.
The above figure shows the percentage of missing weekly death values by cause per state ordered by population (greatest to least). The upper-right corner of the figure are smaller population states with 30% or greater of missing weekly death values for many of the cause of death categories. Insights for these states’ categories will be heavily biased and should be avoided.
It is interesting that 10 out of 12 smallest population states have weekly death counts from 1 to 9 for the diabetes category more than 87% of the time. States with greater populations than Idaho, have weekly counts of diabetes death less than 25% of the time with the largest states seeing diabetes deaths every week.
Another interesting finding is that California had at least one week where there were only 1 to 9 sepsis deaths for the state.
The next figure is the aggregated state data correlation matrix for the causes of death categories. As mentioned previously, the larger states have a greater influence on state aggregated data due to fewer missing values.
State Cause of Death Correlation
Looking at the scatter plot results, there are few linear models that represent the aggregated data. The highest correlations are with natural causes and all causes which show an almost perfect positive correlations with heart, brain and cancer (malignant) deaths.
Cancer has three clusters of data separated by voids. I expect these are due to population differences. Abnormal findings followed by sepsis have the least significant correlation.
Most of the distributions are bi-modal and multi-modal with right skew. This is also likely to population differences.
The next figures will examine the clusters to determine their origin.
State Distribution Analysis
The above figure of the correlation matrix for natural cause, sepsis and malignant examines the clusters in the scatter plots. The clusters appear between 700 to 1,000 and 1,100 to 1,200 on the malignant deaths scale. Next, we will compare each state’s chronological malignant weekly deaths to confirm these are related to population differences.
The above figure is a scatter plot of all states’ malignant causes of death. The states are ordered in the legend by population size, greatest to least. The purpose of the plot is to examine the cause of the clusters in the malignant data. It is confirmed the clusters are due to more populous states with the first cluster being from California’s cancer deaths followed by Texas and Florida combining around 800 cancer deaths.
It is interesting that New York is 4th in population but shows 7th for cancer deaths and is below the smaller population states of Pennsylvania, Illinois and Ohio. I would think a dense city with older buildings and waste systems would have a higher incident of cancer deaths. Some possible explanations may be the state’s citizens spend less time outdoors in the sun, live healthier life styles (non-smokers) or possibly New York had better outcomes when treating cancer. Comparing states’ data on cancer diagnosis versus cancer deaths could be an area of future study to identify possible explanations for the lower cancer death rate.
The next figure looks at the distribution for each state for all_causes and then looks at the distributions for flu and pneumonia, and abnormal findings to examine how the missing values affect the distributions for each state.
All state distributions for the all_causes category look normally distributed to some degree with most having a right skew due to 2017, 2018 and 2019 deaths. This is likely due to increases in population but would require population data to confirm. If population is the cause for the increases, the pattern will continue causing the mean to shift right over the long term.
Next, are the distributions for flu and pneumonia deaths for each state.
The distributions for flu and pneumonia are possibly normal for more populous states but gaps start to show for the missing data beginning with Colorado. Using interpolation looks like it might fix the graphs for Colorado, Oregon, Utah, Oklahoma and a few others states. Though, for many states after West Virginia, interpolation would not be reliable due to too few values existing, causing the state data to go multiple weeks without values. Due to these gaps in data, the analysis of state cause data will be limited to states’ weekly death category data that has no more than 30% missing values.
Abnormal finding shows possible normal distributions for the first 12 most populous states and then sporadically for other states such as Minnesota and Alabama, for least populous states. After Oregon, the missing data is too great to determine a distribution without significant interpolation. This would lead to unreliable insights. Therefore, analysis will be limited to states’ weekly death category data that has no more than 30% missing values.
The next table examines the weekly death categories with the most significant growths.
State Most Significant Changes
There are 43 instances where the 2019 increase in the cause of death has a 95% or greater significance based on a one-tail test. The most significant state and category is Texas for diabetes in 1st and 2nd quarters of 2019.
The next table gives the states and weekly death categories with the least significant growths.
The least significant causes of deaths is flu and pneumonia for Wisconsin during the first quarter of 2019. For the 10 lowest significant causes and states, flu and pneumonia appears 9 out of 10 times for the 1st and 4th quarters (Winter and Fall).
It is interesting that of the 9 states listed in the lowest significance for flu and pneumonia, all but Louisiana have freezing winter temperatures. Yet, these states have the smallest changes in flu and pneumonia deaths. This could be due to consistent year over year flu and pneumonia deaths, the states health care services are better prepared to deal with flu and pneumonia deaths, the populations being more cautious and adhering to preventative measures, or some other cause.
The following figures shows occurrences of causes of deaths over 95% significance by state, cause of death and quarter.
State Most Significant Increases by State, Cause and Quarter
Seven states had occurrences greater than the 95% level of significance (2 standard deviations). Half of the states are in the top 10 most populous states.
Diabetes has the highest occurrence of weekly deaths above the 95% significance level followed by the more general categories of natural and all causes. There were 19 occurrences of causes being over the 95% significance level.
There are 19 instances in 2019 where a disease category had a quarter with a greater than 95% significance level. Most instances occurred during the 2nd quarter followed by the third. This indicates that the most significant changes in deaths occurred during the spring and summer. Though, there are three occurrences in the 1st and 4th quarters (Fall and Winter).
The following figures show the yearly data for the above 95% significance data by each state and cause.
State Highest Significance Causes Year Over Year Trends
In the figures above, there is a general upward trend in causes of death categories for:
- Florida for cerebral (brain)
- Missouri for Alzheimer
- New York for other respiratory
- North Carolina for all causes and natural causes
- Tennessee for natural causes and
- Virginia for all causes and natural causes
New York had an increase in abnormal causes in 2016, was stable in 2017 and 2018 and then had an increase in 2019.
Diabetes in Texas was stable for years 2014 through 2016 and then increased the following years with the greatest increase occurring in 2019.
The next figures compare the quarterly data for 2019 compared to the previous years for each state, cause and quarter that were over the 95% significance level.
State Statistical Data: 2019 to Previous Years
## [1] "The Florida data shows a significant increase in cerebral deaths for the 3rd and 4th quarters of 2019 with both quarters greater than 2 standard deviations."
## [1] "The Missouri data has a significant increase of greater than 2 standard deviations for Alzheimer deaths for the 3rd quarter of 2019."
## [1] "New York had significant increases for abnormal causes for the 1st, 2nd and 4th quarters of 2019."
## [1] "North Carolina had significant increases for all causes, natural causes and week sum for the 2nd quarter of 2019. All causes and natural causes are over 2 standard deviations."
## [1] "Tennessee had significant increases for all causes, natural causes and heart disease deaths for the 2nd quarter of 2019. All causes is above 2 standard deviations"
## [1] "Texas had significant increases for diabetes for the 1st, 2nd and 3rd quarters of 2019. The 1st and 2nd quarters were over 2 standard deviations."
## [1] "Virginia had significant increases for all causes and natural causes for the 3rd quarter of 2019. Natural causes is above 2 standard deviations."State Most Significant Analysis
Texas shows a significant increase in Diabetes deaths compared to other years. The changes from previous years show that there are not week lower than 2018 and most weeks are have 10 or greater deaths than 2018.
The Florida data shows an increase in the number of cerebral deaths for the 2nd half of 2019 when compared to 2018. Though, when comparing the all of 2019 to all of 2018, there seems to be a similar increase in deaths. The most significant increase in cerebral deaths occurred in 2015 and may be due to a low cerebral death rate in 2014.
New York’s increase in other respiratory deaths seems similar to 2017 and 2018 and is likely not a significant increase.
The increase in New York’s abnormal findings it much larger in 2019 than in the previous two years. The 2016 increases look similar but with a lesser smaller increase in the amount of deaths due to abnormal findings for the entire year.
The Missouri data for Alzheimer looks similar to other years with possibly a significant increase during the 2nd half of 2019. The 2019 year started with the most significant decrease in Alzheimer deaths when compared to all other years.
North Carolina’s plots for the all causes categories show similar patterns as previous years and is not likely a significant increase.
Tennessee’s all causes category 2019 data looks similar to previous years with the exception of a large decrease at the beginning of 2019. There is a large increase at the beginning of 2018 likely due to the notable increase in flu and pneumonia deaths that year. The other categories flagged as significant show a similar pattern to the all causes plot and are likley not significant.
Virginia’s all causes category 2019 data looks similar to Tennessee’s data. The whole year data for 2019 is also similar to previous years trends. The exception is that most weekly increases in deaths occurred during the Spring and Summer months. All other years show small increase during this period of time. Virginia may be significant for this range.
Final Plots and Summary
The above figure shows the US cause of death trends on a weekly basis from 2014 to the end of 2019. The magnitude order of the causes of death have remained the same except for flu and pneumonia deaths which has peaks every January.
Seasonal trends can be seen in the figure with Fall and Winter seasons (dark background regions) having peaks and Spring and Summer months having valleys. The consistent peaks and valleys across all diseases suggest a possible relationship between the seasons and possibly between diseases. The long-term growth rate trend of cerebral (brain) weekly deaths appears to be greater than lower respiratory weekly deaths and will likely place it as the third highest cause in the near future.
The correlation between the cause of death categories show moderate to strong relationships for the majority of the categories. This was found to be due to seasonal patterns in the data. The lowest correlation coefficient values occur between abnormal findings and the other categories. The strongest correlations are between heart disease and the other categories.
These correlations show possible relationships and not likely causality. The seasonal pattern is likely the result of people with underlying conditions having lowered immune system responses and being more susceptible to viruses that spread in the winter months. Further study and additional data would be needed to find the cause of the pattern.
The figures above show the US weekly deaths from flu and pneumonia by year and the changes in the weekly death from the previous year. During the Spring and Summer seasons, flu and pneumonia weekly deaths have remained steady with even slight decreases over the years. The weekly death counts begin increasing around week 45 with peaks in the 1st quarter of the next year.
An interesting trend seen for these six years is the peaks seem to have greater maximums when they occur within the first two weeks of the new year. If they occur after the 5th week, the peaks’ maximums tend to be 50% lower. There seems to be a relationship between the high and low peaks. For example, the 2015 plot had the 2nd highest peak and the following year had the lowest peak. The same pattern occurs with 2018 which had the highest peak in the 1st quarter of the year. The following year was the 2nd lowest peak. This could be due to the population being more cautious following a high flu death year. It also could be due to vaccines not being as effective one year and being improved for the next.
The figure on the right shows the increases and decreases in flu and pneumonia weekly death counts compared to the previous year’s week. The figure shows that flu and pneumonia deaths are typically decreasing year over year except for the spikes seen at the beginning of 2015, 2017 and 2018. It is notable, that following the two greatest spikes in 2015 and 2018, there are even greater decreases the following year.
The above figures compare Texas’ 2019 diabetes deaths to all previous years deaths. It shows statistically significant increases for every quarter of 2019 with the 1st and 2nd quarters being greater than 2 standard deviations compared to the previous years means.
The next figure takes a look at the week by week data for each year to see the affect of a year with low weekly death counts pulling the other years’ means down.
The 2019 year’s data shows a significant increase in weekly Texas diabetes deaths when compared to previous years. The figure on the left shows a generally larger area between the 2019 line and the 2018 line indicating a large increase in weekly deaths from diabetes. The significance of this increase in weekly deaths are seen in the large yellow area on right indicating a large increase in diabetes deaths during every week of 2019. The majority of weeks in 2019 had 10 or more additional deaths from diabetes with some weeks having over 30 additional deaths.
Reflection and Areas of Further Study
This report explored the CDC Weekly Morbidity and Mortality data for years 2014 through 2019 which contains 16,902 observations and 30 variables. This data set was chosen to answer questions about causes of deaths in the United States prior to Covid-19. Separating the data into Unites States deaths and state deaths allowed the most significant increases to be examined at a national and state level identifying general trends versus local trends.
The state level data was only partially analyzed due to missing values for less populous states. It was found the less populous states had many death counts in the 1 to 9 range which are redacted values from the data due to privacy concerns. Though interpolation could be used on some of the smaller states’ missing data, this would introduce bias and reduce confidence in the findings. Therefore, state categories that were missing 30% or more of the values were omitted from the analysis.
The results of the United States data showed an increasing trend in weekly death rates for all causes, natural causes, heart disease, cancer, lower respiratory, cerebral, Alzheimer’s, diabetes, renal (kidney), other chronic respiratory and abnormal findings cause of death categories. Flu, pneumonia and sepsis show declines over the years. The most significant increase of 2019 compared to the aggregated previous years’ mean was diabetes.
Examining the state data showed that the most significant increase in cause of death categories was diabetes in Texas during the 1st quarter of 2019. Examining the remaining quarters of 2019 for diabetes deaths in Texas, it was found the entire 2019 year had significant increases in diabetes deaths compared to the combined mean of the previous year. The 1st and 2nd quarters showed more than a 2 standard deviation increase. Other categories with significant increases were found to be due to the distributions of previous years data and not likely statistically significant. An exception, is New York which showed possible statistical increases for the abnormal findings category during the 1st and 2nd quarters of 2019.
Areas for future study
Further studies could include the effects of Covid-19 on weekly death categories. By combining the 2014-2019 data set with the 2020-2021 data that includes Covid-19 deaths, when available. A question that could be answered is, how do seasonal viruses effect weekly deaths in all categories? It would be interesting to see if the seasonal patterns found in the data disappear when preventative measures such as distancing, masks and frequent hand washing was practiced. The combined data may even be used to find anomalies in the 2019 data that may have indicated Covid-19 was spreading locally before Covid-19 testing was available. One such example relates to E.V.A.L.I. (Vape Lung) deaths in the summer and fall of 2019. EVALI has similar symptoms as Covid-19 and is a diagnosis of exclusion. E.V.A.L.I was never found in any other country. If would be interesting to see if E.V.A.L.I diagnosis disappeared once Covid-19 tests were available.
Limitations of the analysis
Limitations in the data were the effect of missing data, voluntary reporting of data and possible reporting and sampling biases. States could not be compared to each other due to participation rates in reporting varying by state. In addition, one state may report 30% of the deaths for a specific cause while another may report 15% of the deaths for the same cause. Additionally, causes of death where multiple causes are a factor could lead to a bias towards one diagnosis over another.